I would like to send out a huge THANK YOU to anyone who interacted or followed the Shaving Points College Basketball Model over the past four months. It was something that I created on a whim back in November, and never imagined it would have gained the support it did.
I started building it simply as a challenge to myself to see if I could effectively gather and write a code for the mass amounts of data I deal with in CBB. I was skeptical that I could even get it to process the data, let alone pick winners.
During the off season I aim to continue improving the model’s efficiency, accuracy, and readability. At this point, I want to give you all a more in depth run down of the model’s calculations. Below are the factors and weights for each factor:
1. Handicap Model #1
The model uses advanced stats to generate a handicap for each game. the calculations are very similar to those found in this post.
The stats used to generate my own handicap include TO rate and TO rate allowed, Effective FG % for and allowed, offensive and defensive rebound rates, FT rates for and allowed. For more details on this section, I suggest going to the post 7 years ago by u/DerekJohn. His post (linked above) does an outstanding step-by-step guide. I first started playing with this model system 3 years ago and have been taking it ever since. It’s a wonderful starting point.
2. Handicap Model #2
Diving into the second handicapping model. I made a simple calculation using offensive rating and defensive rating per possession, and each teams’ average possessions.
3. Creating the Baseline Confidence
I use the difference between each projected spread and the real spread to set a baseline confidence for each team to cover. In order to do this, I divide the differential by the expected total points in the game. This is what I call the confidence index, which I add to 50%. If our initial baseline is that each team has a 50/50 chance before looking at stats, this confidence index is the difference between the two teams season stats.
4. Factoring for Season Results ATS
Next, I factor in each team cover % ATS during the season at a weight of 1/3. I also factor in their home / road cover % at a weight of 1/3, and their KenPom SOS & Luck ratings (a simple combination of the two stats) at a weight of 1/3. Adding each of these factors at those weights generates the confidence %.
The moneyline confidence % is simply the result of setting the spread to zero in the model.
The NBA model is very similar but includes adjustments for injured players. That has really struggled since the trade deadline, as a lot of factors changed, and full season stats were no longer as relevant to current expectations.
Thank you all again and be on the lookout for an improved model next season.